JP5719564B2 - Apparatus and method for accurate ECG measurement during cardiopulmonary resuscitation with segmented stitching adaptation algorithm - Google Patents

Description

  The present disclosure relates generally to the field of systems, methods, and devices for processing electrocardiogram (ECG) signals. More particularly, the present invention relates to a system, method and apparatus for adaptively removing artifacts in ECG signals resulting from cardiopulmonary resuscitation (CPR).

  Almost 20 years have passed since the development of automatic external defibrillators (AEDs) to help reduce cardiac arrest accidents. During that period, AEDs have become widespread in public areas such as offices, shopping centers, stadiums, and other places where traffic is heavy. The AED empowers citizens to provide medical rescue during cardiac emergency situations in public places that were not previously rescued during the critically early stages of a cardiac event. Particularly in recent years, fully automatic external defibrillators that can accurately detect ventricular arrhythmias and non-shock supraventricular arrhythmias have been developed to treat patients without attendants (for example, Patent Document 1). reference). These devices treat patients suffering from ventricular arrhythmias and have high sensitivity and specificity in detecting shock arrhythmias in real time. Furthermore, AED has been developed to work as a diagnostic monitoring device that can automatically provide treatment in a hospital (see, for example, Patent Document 2).

  In addition to developments in the AED field, there have been several advances in understanding the link between human physiology and medicine. These advances in medical research have led to the development of new protocols and standard operating procedures in dealing with physical trauma accidents. For example, in public access protocols for defibrillation, recent guidelines have emphasized the need for cardiopulmonary resuscitation (CPR) along with the use of AEDs. In fact, the recent American Heart Association (AHA) cardiopulmonary resuscitation and emergency cardiovascular medical guidelines have shown that an AED can detect a shock rhythm and then prompt the rescuer to restore pressure immediately. It is suggested that it can be incorporated into a corresponding protocol (for example, see Non-Patent Document 1). In addition, guidelines have been developed by the AED, specifically reducing or reducing the number of pending pressures through patient re-evaluation and further re-educating or supporting rescuers in the direction of ensuring efficient transfer to experienced medical professionals. Explains that it can be done. The guidelines, along with their own research, have resulted in a comprehensive method with defibrillation, along with CPR, as a suggested use for AED devices.

  Current AEDs do not work in performing the current proposed use of AEDs as recommended by the guidelines while providing defibrillation. Most AEDs available today attempt to classify ventricular rhythms. Specifically, current AEDs attempt to distinguish between shocking ventricular rhythms and all other non-shock rhythms. This detection and analysis of ventricular rhythm requires real-time analysis of the ECG waveform. Thus, the functionality, accuracy, and speed of AEDs are highly dependent on the algorithms and hardware used for real-time analysis of ECG waveforms.

  In many implementations, the algorithm is based on heart rate calculations and various morphological features derived from ECG waveforms such as ECG waveform factors and irregularities as disclosed in US Pat. Dependent. In order to provide more processing power, current AEDs typically incorporate algorithms and control logic within the microcontroller.

It is known that current algorithms and specialized hardware implementations can have a serious impact on the effectiveness of the AED. Specifically, the signal-to-noise ratio of the ECG signal is AED
It greatly affects the performance. For example, during rescue operations, many current AED-implemented algorithms require several seconds of clean ECG signal data to classify detected ventricular rhythms. The opportunity for rescuers to obtain such clean signal data is significantly reduced during cardiopulmonary resuscitation where chest compressions and relaxations can be applied at a defined rate approaching 100 cycles per minute. In practice, chest compressions and relaxations generate significant motion artifacts in ECG recording. In addition, ECG signals exhibit poor amplitude during ventricular arrhythmia events and further reduce the signal-to-noise ratio, often resulting in poor quality or unusable signals. In these situations, existing arrhythmia recognition algorithms may not function properly, putting the victim at risk.

  Attempts have been made to reduce the effects of sensory artifacts by changing the design of ECG electrodes and analog front-end circuitry. One design implements a low cut-off frequency to achieve a high cut-off in the ECG amplifier. Other designs utilize differential amplifiers with very high common mode rejection ratio (CMRR) in an attempt to avoid some artifacts. However, in these designs, it is essential to capture good quality signals in the digital domain in order to remove any artifacts using digital logic and algorithms. This is mainly due to the fact that the signal quality which has been degraded due to saturation effects during the conversion from analog to digital cannot be recovered using currently known techniques.

  In addition to electrode design, current algorithms are ineffective in filtering CPR current standards and practices. One of the challenges is to identify shock heart rhythm despite the CPR compression cycle and to identify non-shock / demodulation rhythm in real time. Since cardiac arrest is an important metric, another challenge is to accurately detect cardiac arrest. Various methods have been proposed for identifying and removing CPR artifacts that can disrupt the ECG signal. For example, Patent Document 4 uses a reference signal in an attempt to remove artifacts. U.S. Patent No. 6,057,028 provides an algorithm that relies on assumptions regarding the operation of the heart system along with a reference signal. U.S. Pat. No. 6,057,096 uses a number of sensors as an indicator of CPR activity and to provide a reference signal. U.S. Pat. No. 6,057,059 utilizes a reference signal indicative of CPR activity to identify the presence of CPR artifacts in an ECG segment. U.S. Patent No. 6,057,836 discloses the use of transthoracic impedance measurement to identify significant patient movement. Patent Document 8 describes using a plurality of ECG leads to estimate a noise effect on an ECG signal. Japanese Patent Application Laid-Open No. 2004-228561 discloses a frequency domain method for identifying a system using linear predictive filtering and a recursive least square method. Several recent studies have introduced alternative methods of ECG filtering based on adaptive regression on delayed reference signals (see, for example, Non-Patent Document 2). Yet another method of CRP artifact detection and filtering focuses on the use of frequency modulation instead of a reference signal to remove anomalies (see, for example, Non-Patent Document 3). Other disclosed methods of performing therapy in response situations focus on the detection and measurement of CPR activity, and a chest compression detector (see, for example, US Pat. Patent Document 11) is used.

US Pat. No. 5,474,574 US Pat. No. 6,658,290 US Pat. No. 6,480,734 US Pat. No. 6,961,612 US Pat. No. 7,039,457 US Pat. No. 6,807,442 International Publication No. 2006/015348 Pamphlet US Pat. No. 5,704,365 US Pat. No. 7,295,871 European Patent Application No. 1859770 US Pat. No. 7,122,014

American Heart Association Guidelines for Cardiopulmonary Resuscitation and Emergency Cardiovascular Care, (USA), IV-36, American Heart Association, Inc., 2005. Rheinberger et al., “Removal of resuscitation artifacts from ventricular fibrillation ECG signals using kalman methods” (USA), Computers in Cardiology, 2005. Aramendi et al., “Detection of ventricular fibrillation in the presence of cardiopulmonary resuscitation artifacts” (USA), Resuscitation, 2007.

  However, all of these platforms or methods have limitations and problems when providing real-time treatment in accordance with recent American Heart Association CPR guidelines. Therefore, a method and apparatus for filtering CPR artifacts from an ECG signal that is effective across various ECG segments, showing low computational cost, and showing near real-time analysis and filtering, thus determining shock and non-shock conditions A method and apparatus that enables as clean a signal as possible is desired.

  Various embodiments of the present invention disclose methods and apparatus for filtering signal artifacts from a sensed ECG signal in real time. Various embodiments have an apparatus or automated method that utilizes a piecewise stitching adaptation algorithm to filter signal artifacts from an ECG signal. Various embodiments implement a piecewise stitching adaptation algorithm in computer hardware such as a specially designed computer processor or microprocessor. Other embodiments store the segment stitching adaptation algorithm in a non-volatile computer accessible memory. In various embodiments, the hardware is connected to sensors that sense physiological signals. In certain embodiments, one of the sensors detects an ECG signal. In other embodiments, other sensors detect the artifact signal. The artifact signal can be a CPR compression signal, hemodynamic signal, or other signal that reflects yet another physiological function that can cause artifacts in the ECG signal. Further, representative signals of CPR artifacts can be obtained by using sensing techniques such as ultrasound, light detection, and cardiograms that can indicate the physical cause of the artifact.

  The method and apparatus may then execute a piecewise stitching adaptation algorithm to remove artifacts caused by the artifact signal from the ECG signal by selecting a signal sample window from the ECG signal and the artifact signal. A primary signal segment and a secondary signal segment can then result from the ECG signal and the artifact signal. The relationship between the primary signal segment and the secondary signal segment can then be measured. This may allow estimation of signal artifacts in the primary signal based on the measured relationship. Finally, various embodiments may remove the estimated signal artifact from the primary signal segment.

  In various embodiments, shock ECG rhythms can be identified by utilizing a rhythm analysis algorithm. This may allow the method and system to be utilized in medical devices such as automated external defibrillators (AEDs). Rhythm analysis algorithms may enable the implementation of life support treatments that meet recent operational, clinical, and CPR guidelines.

  In various embodiments, the present method and apparatus will optimize filtering and detection only by performing an artifact filtering process when an artifact signal is detected. In this way, the power and latency required for therapeutic application is reduced. In other embodiments, the signal sample window is selected from ECG signals and artifact signals in signal sample windows of uniform size and non-uniform size depending on the time delay between the ECG signal and the artifact signal. Thus, the method and apparatus can handle detection delays due to perceptual deficits and delays resulting from physiological function differences. In certain embodiments, the start and end times of signal segments in the ECG and artifact signals will be comparable. In other embodiments, the ECG signal and artifact signal sample window start and end times will be determined by utilizing adaptive indexing or segmented regression.

  In various embodiments, the piecewise stitching adaptation algorithm will estimate the phase lead or phase lag between the ECG signal and the artifact signal by using a shift autocorrelation calculation. These estimates can then be stored in memory for future use in selecting further signal sample windows. In various embodiments, the piecewise stitching adaptation algorithm utilized a weighting method that applies weights to the primary and secondary signal segments. In certain embodiments, all segments are equally weighted. In other embodiments, center segment weighting is utilized to provide greater weight to the center signal segment.

  In various embodiments, the method and apparatus may detect artifacts that occur in other signals that are indicative of other physiological processes. Thus, in certain embodiments, the artifact signal is a measure of hemodynamic activity. Further, the method and apparatus may utilize passive or active filtering on the primary signal in providing the artifact signal. Thus, in certain embodiments, the ECG signal is filtered by using a bandpass filter to provide an artifact signal. In these embodiments, the method and apparatus require only one sense signal to filter signal artifacts.

Schematic diagram of an automatic external defibrillator. 1 is a schematic diagram of an automatic external defibrillator that utilizes a segmented stitching adaptation algorithm in accordance with one embodiment of the present invention. FIG. 4 is a graph of a piecewise stitching adaptation algorithm utilizing convolution, according to one embodiment of the present invention. 6 is a flowchart illustrating the operation of a piecewise stitching algorithm that utilizes convolution, in accordance with one embodiment of the present invention. FIG. 4 is a graph of a piecewise stitching algorithm that utilizes regression, in accordance with one embodiment of the present invention. FIG. 6 is a flowchart illustrating the operation of a piecewise stitching algorithm that utilizes regression, in accordance with one embodiment of the present invention. FIG. 4 is a graph of a piecewise stitching algorithm that utilizes variable window regression, in accordance with one embodiment of the present invention. FIG. 6 is a schematic diagram of an implementation of a piecewise stitching algorithm, according to one embodiment of the present invention. FIG. 6 is a schematic diagram of an implementation of a piecewise stitching algorithm, according to one embodiment of the present invention. FIG. 6 is a schematic diagram of an implementation of a piecewise stitching algorithm, according to one embodiment of the present invention. FIG. 6 is a schematic diagram of an implementation of a piecewise stitching algorithm, according to one embodiment of the present invention. 1 is a schematic diagram of a segmented stitching algorithm incorporated into a defibrillator, in accordance with one embodiment of the present invention. Graph of ventricular tachycardia waveform damaged by CPR artifact. Damage and noise waveform graph. FIG. 6 is a graph of a ventricular tachycardia waveform recovered using a segmented stitching algorithm in accordance with one embodiment of the present invention. 1 is a schematic diagram of an automatic external defibrillator that utilizes a segmented stitching adaptation algorithm in accordance with one embodiment of the present invention.

  As mentioned in the background, several algorithms have been implemented in current AEDs in an attempt to meet the revised AHA guidelines for cardiopulmonary resuscitation and emergency cardiovascular therapy. One implementation is an adaptive filter technique, which is briefly outlined herein in connection with the general schematic of an AED that incorporates an algorithm as provided in FIG.

Adaptive filters can be implemented using several algorithms, but the least squares (LMS) algorithm and its derivatives are most often used. In the LMS adaptive filter, a mean square cost function (ie) ζ = E [e ² (n)] is assumed. The adaptive filter then minimizes the instantaneous square error, ζ (n), by using the steepest gradient algorithm. This algorithm updates the coefficient vector in the negative gradient direction with step size μ. For example, in the case of an FIR adaptive filter:

w (n + I) = w (n) −μ2. N′ζ (n) (A)
In the equation, the weight w (n) can be adjusted for each sample. Another algorithm utilized in many fitting algorithms is the recursive least squares (RLS) algorithm. In the RLS algorithm, the cost function is given by

計算的に重み｛ｗ（ｎ）｝の値の更新は、ＬＭＳとＲＬＳに基づく両適応フィルタに対して試料ごとに行われなければならない。これらの計算は、非常にコストがかかり、多重計算が更新ごとに必要とされる。更に窓ごとに全ての試料に行われている計算を調整する方法は無い。加えて、適合アルゴリズムは、整定時間を有し、最小誤差出力または有用信号（ノイズが除去された）の整定時間は、数秒かかる。整定時間はまた、重みの初期値と、ＲＬＳアルゴリズムのパラメータλとＬＭＳアルゴリズムの刻み幅μとに依存する。

The computational update of the value of weight {w (n)} must be done for each sample for both adaptive filters based on LMS and RLS. These calculations are very costly and multiple calculations are required for each update. Furthermore, there is no way to adjust the calculations performed on all samples for each window. In addition, the adaptation algorithm has a settling time, and the settling time of the minimum error output or useful signal (with noise removed) takes several seconds. The settling time also depends on the initial value of the weight, the parameter λ of the RLS algorithm and the step size μ of the LMS algorithm.

In the CPR artifact removal problem, time-varying ECG and CPR signals introduces complexity to the adaptation process. ECG signals vary from ventricular tachycardia (VT) to ventricular fibrillation (VF), low amplitude ventricular fibrillation, and cardiac arrest. Furthermore, the recovery signal also varies from cardiac arrest to low amplitude ventricular fibrillation, ventricular tachycardia, supraventricular tachycardia (SVT), and the like. The frequency and amplitude of all of these waveforms show huge fluctuations within the waveforms within themselves. In addition to these variations, CPR artifacts such as compression and inflation vary widely between rescuers, as well as during specific cycles of CPR. Furthermore, the actual amplitude of compression and expansion varies greatly. In essence, the variability of ventricular signals consistent with CPR artifacts interferes with the settling of the adaptive filter, thus reducing the computational power of the device implementing this method.

In view of the capabilities and shortcomings of this adaptive filter technology today, embodiments of the present invention will now be described.
Other embodiments provide a solution for removing noise in ECG signals using a piecewise stitching adaptation algorithm (PSAA). In various embodiments, the piece stitching adaptation algorithm may utilize at least one of a piecewise regression method and a piecewise deconvolution method to efficiently analyze and clean the sensed ventricular signal.

  In various embodiments, the segment stitching adaptation algorithm utilizes a reference signal received from a device that measures CPR activity. These embodiments allow the segment stitching adaptation algorithm to have a baseline or reference for all CPR activities. By using the CPR baseline or criteria, it is then possible to correlate CPR activity to the detected ECG. The acquisition techniques used to obtain the CPR signal, ie source, sampling technique, filtering method and sensor, reflect what is used to obtain the ECG signal. For example, the common mode ECG can be used to measure CPR reference signals to ensure a proper representation of CPR activity. By utilizing the same technique, the CPR reference signal can be mapped one-to-one with the ECG signal reference. Thereby, an instantaneous correlation between ECG data and CPR data occurs. Other embodiments may utilize reference signals generated from mechanical acceleration, velocity, or distance measurement sensing. However, in the use of these other reference signals, the accuracy may be reduced due to possible causal relationships between the ECG reference signal and the artifact components. Thus, embodiments may utilize time series CPR reference signals along with time series ECGs to increase accuracy.

  By using the same technique in measuring the CPR reference signal and the ECG signal, accuracy is increased, but the detection signal may still not be aligned 1: 1 in the sample frame. This may be due in part to the propagation of mechanical or electrical signals through various body tissues. For example, the signal can be deformed by conducting through muscle cells. In other situations, the method utilized to detect the ECG and CPR signals may be slightly different due to differences in sensory device or other system requirements. However, to a slight extent, these differences can also introduce delays and thus lack a one-to-one correlation. Various embodiments may utilize a convolution function or a transfer function to assist in measuring the relationship between the CPR signal and the ECG signal. This makes it possible to correlate multiple samples of one time series with multiple samples of another time series. For example, more than one sample segment in the artifact component of the ECG signal can be associated with a similar large segment in the CPR reference signal. Once established, artifacts may be removed from the ECG signal by utilizing the relationship between the CPR reference signal and the ECG signal. In various embodiments, both the instantaneous segregation algorithm and the deconvolution algorithm are utilized to remove CPR artifacts from the ECG signal.

  In order to understand the present invention, some considerations of the various signals and the principles involved in all of the ECG and CPR signaling must be discussed. By understanding these principles and how these principles relate, various embodiments of the invention can be more clearly understood.

Signals such as ECG signals, artifact component signals that affect ECG signals, and reference signals, for example, are considered probabilistic or random signals. Such a signal cannot be arbitrarily reproduced. The main statistical parameters representing a stochastic signal are its mean, variance, and autocovariance.

  Actual signal processing or time series estimation is possible only when the signal exhibits ergodic properties. A stochastic signal is defined as an ergodic signal if all of its statistical properties can be estimated from sufficiently large finite lengths, respectively, realized. For an ergodic signal, as the implementation length goes to infinity, the time average is equal to the ensemble average obtained by the expectation operator in the limit.

  The estimation formula for the actual ergodic signal is shown below.

上記の限界演算の代わりに有限和が、以下に示されるように用いられることが可能である。

Instead of the above limit operation, a finite sum can be used as shown below.

ランダム信号に対しては、自己相関、または自己共分散関数は、極めて重要な役割を果たす。仮に、ｅｃｇ信号｛ｅｃｇ（ｎ）｝が、｛ｓ（ｎ）｝と｛ｒ（ｎ）｝の重ね合わせからできている場合（｛ｓ（ｎ）｝は、クリーン信号成分を表わす。｛ｒ（ｎ）｝は、ランダムノイズ成分である）、その自己相関は、以下のように表わされることが可能である。

For random signals, the autocorrelation or autocovariance function plays a very important role. If the ecg signal {ecg (n)} is composed of a superposition of {s (n)} and {r (n)} ({s (n)}) represents a clean signal component. (N)} is a random noise component), and its autocorrelation can be expressed as:

E {ecg [n] ecg [n + l]} = E {(s [n] + r [n]) (s [n + l] + r [n + l])}
= E {s [n] s [n + l]} + E {s [n] r [n + l]} + E {r [n] s [n + l]} + E {r [n] r [n + l]} (7).

Since {ecg [n]} and {r [n]} are not correlated, Equation (7) is reduced as follows.

E {ecg [n] ecg [n + l]} = E {s [n] s [n + l]} + σ ² _r
(8)

Thus, autocorrelation preserves the signal content, but limits random uncorrelated noise to the dc component in many practical situations. Thus, autocorrelation and cross-correlation are used in a random signal estimation task to analyze random signal characteristics and their interaction with another signal. This is superior to using raw signals to estimate random signal characteristics as analysis.

With this knowledge, the invention described herein can be clearly understood. Assume that the observed ECG signal during cardiopulmonary resuscitation (CPR) is {y (n)}. The {y (n)} is obtained from a combination of the above-described ECG signal component {ecg (n)} and non-correlated wideband noise {N ₁ (n)}, and an artifact component {a ₁ (n)}. Composed. Let CPR reference signal indicating CPR activity be {x (n)}. Mathematically, this relationship can be expressed as:

{Y (n)} = {ecg (n)} + {a ₁ (n)} + {N ₁ (n)}.
{X (n)} = {b (n)} + {a ₂ (n)} + {N ₂ (n)} (9)

As noted above, in various embodiments, {y (n)} refers to the ECG signal observed when recorded from an automated external defibrillator electrode, and {ecg (n)} is true {A ₁ (n)} denotes an artifact component observed in the ECG observed when CPR is performed. {N ₁ (n)} and {N ₂ (n)} denote uncorrelated wideband noise that is always present in any electronic sensor system. If CPR is not performed, the artifact component {a ₁ (n)} should be zero. In another embodiment, {x (n)} indicates a CPR reference signal indicating CPR activity. The {x (n)} is the baseline activity component {b (n)} that should be very close to zero in all situations, the real artifact signal {a ₂ (n)} recorded by the CPR sensor, and the uncorrelated wide area It is composed of band noise {N ₂ (n)}. In various embodiments, {x (n)} is zero if no CPR activity occurs. Thus, in various embodiments, the artifact {a ₁ (n)} is estimated and removed in {y (n)} by using {x (n)} so that a clean ECG signal {ecg ( n)} occurs.

  An additional constraint in various embodiments is to limit the implementation of the entire operation in the time domain. Windowing the data into a number of small windows provides the opportunity for the AED to perform this analysis in a live rescue operation as it provides an opportunity to implement a real-time algorithm.

E | x (n) x ^T (n) | = R _xx (0) = x autocorrelation and E | x (n) y ^T (n) | = R _xy (0) = cross-correlation between y and x (10)

If the above two relationships are further expanded, the following relationship is obtained.

つまり、上記の計算は、自己相関（ＡＣＳ）と相互相関（ＣＣＳ）数列に導く計算の部分集合である。

That is, the above calculations are a subset of calculations that lead to autocorrelation (ACS) and cross-correlation (CCS) sequences.

種々の実施形態は、ＣＰＲ基準信号｛ｘ（ｎ）｝を観測ＥＣＧ信号｛ｙ（ｎ）｝に関連付ける、インパルス応答｛ｈ［ｎ］｝の安定な線形時不変系（ＬＴＩ）離散時間系を利用する。これらの実施形態は、以下の式によって入力−出力関係を定義する。

Various embodiments relate to a stable linear time-invariant (LTI) discrete-time system of impulse responses {h [n]} that associates a CPR reference signal {x (n)} with an observed ECG signal {y (n)}. Use. These embodiments define the input-output relationship by the following equation:

更に、方程式（１２）において定義されるようなＡＣＳが即時計算の範囲内で公知であるという仮説が、これらの実施形態において立てられる。仮説の結果は、方程式（１２）において示されるようなＣＣＳであり、以下のように計算される。

Furthermore, the hypothesis is established in these embodiments that ACS as defined in equation (12) is known within the scope of immediate computation. The hypothetical result is a CCS as shown in equation (12), which is calculated as follows:

（１４）に（１３）を代入すると、以下の方程式が得られる。

Substituting (13) for (14) yields the following equation:

種々の実施形態において長さＮの因果的有限長インパルス応答と方程式（１５）が、以下のように縮小すると考えるのが現実的である。

In various embodiments, it is realistic to think that the causal finite impulse response of length N and equation (15) are reduced as follows:

このようにして、ＡＣＳとＣＣＳ、ｒ_ｘｘ［ｌ］とｒ_ｙｘ［ｌ］の両方が計算される。次に、ＡＣＳとＣＣＳが与えられると、システム同定またはインパルス応答推定が実行される。

In this way, both ACS and CCS, r _xx [l] and r _yx [l] are calculated. Next, given ACS and CCS, system identification or impulse response estimation is performed.

In various embodiments, the recursive relationship for calculating the impulse response sample {h [n]} of the CPR reference signal {x [n]} and the observed ECG signal {y [n]} is ACS {r _xx [l]. } And CCS {r _yx [l]} are given as the same results as the impulse response sample {h [n]}. By using ACS and CCS, the same information as when using the CPR reference signal {x [n]} and the observed ECG signal {y [n]} is provided, but the observed ECG signal {y [n]} It also has the added benefit of reducing the effects of uncorrelated noise.

The following recursive calculation helps to calculate the impulse response sample {h [n]} from the values of ACS {r _xx [l]} and CCS {r _yx [l]}.

従って、アーチファクト｛ａ_１［ｎ］｝は、ＣＰＲ基準信号｛ｘ［ｎ］｝と観測ＥＣＧ信号｛ｙ［ｎ］｝の間の関係から推定または再構築される。次に、アーチファクトは、方程式（９）で示されるモデルにおいて示されるように、クリーンＥＣＧ信号｛ｅｃｇ［ｎ］｝を得るべく、観測ＥＣＧ信号｛ｙ［ｎ］｝から差引かれる。

Accordingly, the artifact {a ₁ [n]} is estimated or reconstructed from the relationship between the CPR reference signal {x [n]} and the observed ECG signal {y [n]}. The artifact is then subtracted from the observed ECG signal {y [n]} to obtain a clean ECG signal {ecg [n]}, as shown in the model shown in equation (9).

In various embodiments, after removal of the estimated artifacts due to CPR, using overlapping windowing results in a continuous output signal {ecg [n]}. Exact overlapping segments can be identified, and contributions for impulse responses {h _i [n]}, {h _{i + 1} [n]}, etc. are evaluated. Proper weighting ensures that there is no spike in the intermediate output, ie the reconstruction artifact component {a ₁ [n]}.

Referring to FIG. 2, an embodiment of a system 200 that utilizes a piecewise stitching adaptation algorithm is shown. In various embodiments, the system may utilize an ECG monitor 202 that receives an ECG signal 204. The ECG monitor 202 may monitor or record ECG signals 204 from sensors installed on the body. The ECG monitor 202 can analyze and adjust the ECG signal 204 for further processing. In addition, the ECG monitor 202 may identify anomalies or artifacts that exist due to additional treatment being performed, such as CPR compression. Various embodiments will have a CPR monitor 206 that analyzes the CPR signal 208. CPR signal 208 may provide an independent indicator of CPR activity being performed. In addition, the CPR monitor 206 may adjust the CPR signal 204 for further processing. The system 200 may then utilize decision logic 210 to determine whether a piecewise stitching adaptation algorithm is required. If logic 210 determines that CPR signal 208 is present, then segment stitching adaptation algorithm 212 will be initialized and invoked. The segment stitching adaptation algorithm 212 will then obtain the ECG signal 204 and the CPR signal 208 and remove the CPR signal 208 from the ECG signal 204 to provide an ECG signal 214 free of CPR artifacts. The ECG signal 214 can then be forwarded to other algorithms in the system 200, such as the arrhythmia detection algorithm 216, for example. In certain embodiments, if the CPR signal 208 is not present, the logic 210 does not initialize the piecewise stitching adaptation algorithm 212 and the system 200 can pass the ECG signal 204 to other algorithms such as, for example, the arrhythmia detection algorithm 216. Will be transferred directly.

  Other embodiments may detect a variety of primary and secondary signals. The secondary signal is selectively removed or filtered by providing additional data or signal input to the primary signal and utilizing a piecewise stitching adaptation algorithm. For example, motion artifacts in loaded ECG test machines, respiratory artifacts in ECG and loaded ECG test machines, tester artifacts in loaded ECG test machines, ECG signals in EEG, ECG signals / pulses from EEG signals, and other hemodynamic signals CPR artifact. Thus, various embodiments eliminate artifacts in various signals representing physical impulses such as ECG, EEG, etc. by utilizing a second signal representing various physical impulses such as CPR compression. obtain.

Referring to FIG. 3, an embodiment of a piecewise stitching adaptation algorithm that utilizes a convolution algorithm is shown. In various embodiments, the overlapping convolution output segments is implemented by first windowing the first 1W _L samples of observed ECG signal {y [n]} and CPR reference signal {x [n]}. W _L indicates the window length. A square window is then applied to both input signals to generate a CPR reference signal segment {x ₁ (n)} and a measured ECG signal segment {y ₁ (n)}. This stage is visualized with the following equation:

y ₁ (n) = y (n) w (n) (0 < n <W _L ) or 0 (others).

x ₁ (n) = x (n) w (n) (0 < n <W _L ) or 0 (others).
(18)

Next, R _xy (k) and R _xx (k) values are calculated for lags up to N _lag points. N _lag is a measure of the time difference between the observed ECG signal segment {y ₁ (n)} and the CPR reference signal segment {x ₁ (n)}. Lead / lag and window estimation is shown, for example, in FIG.

Next, the CCS function and the ACS function are determined as described above in equation (12) after subtracting the average value of {y ₁ (n)} and {x ₁ (n)}. After determining the CCS and ACS functions, {h (n)} is determined from the relationship between the two functions by using equation (17), and the magnitude of {h (n)} is N _lag Set equal. The average value is then removed from the CPR signal by using the following equation:

m ₀ = average {x ₁ [n]}.
_{{X 1 [n]} =} {x 1 [n]} - Mean _{{x 1 [n]} =} {x 1 [n]} - m o (19).

Next, {h (n)} is used to construct {x ₁ ′ (n)} using the following convolution equation. {X ₁ ′ (n)} is an estimated artifact.

信号｛ｘ_１’（ｎ）｝の長さは、Ｗ_Ｌ＋Ｎ_ｌａｇ−１である。

The length of the signal {x ₁ ′ (n)} is W _L + N _lag −1.

The output of convolution _{{x 1} '(n)} is reduced to a 1W _L points. Ideally, the shortening should avoid starting and _ending N _lag points. However, various embodiments only avoid N _lag -1 points. Still other embodiments reduce the impact of the initial N _lag points by giving lower weights in subsequent computations that estimate artifacts. Next, the dc response of the convolution operation is integrated as shown in equation (21).

式中、ｍ_０は、変調比Ｍ（Ｍはｍ／ｍ_０に等しい）における励起の変調幅である。

Where m ₀ is the modulation width of the excitation at the modulation ratio M (M is equal to m / m ₀ ).

Next, the artifact segment {a ₁ (n)} for the data is estimated by utilizing the equation provided in (22).

種々の実施形態において、第１窓と第２窓の間の第１非重複セグメントの出力は、同じである。更に種々の実施形態は、窓がＮ_ｌａｇポイント急上昇すると仮定する。しかし、この急上昇は、入力パラメータによって測定され得るか、でなければＮ_急上昇（＝Ｎ_ｊｕｍｐ）で計算および示され得る。アーチファクトセグメント｛ａ_１（ｎ）｝の測定後、クリーンなＥＣＧ信号ｙ^推定（＝ｙ^ｅｓｔ）が、第１Ｎ_ｌａｇポイントに対して推定され得る。

In various embodiments, the output of the first non-overlapping segment between the first window and the second window is the same. Further, various embodiments assume that the window spikes N _lag points. However, this spike can be measured by the input parameters, or it can be calculated and indicated with an N _spike (= N _jump ). After measuring the artifact segment {a ₁ (n)}, a clean ECG signal y ^estimate (= y ^est ) may be estimated for the first N _lag points.

ＣＰＲ基準信号｛ｘ（ｎ）｝と｛ｙ（ｎ）｝の間で数ポイントだけ一定のリードまたはラグがある場合、両セグメントにおけるＮ_ｌａｇ／２またはＮ_ｌａｇポイントとのシフト相互相関は、特定のリードかラグで最大値を生じさせることになる。上記の減算は、推定アーチファクト信号｛ａ_１（ｎ）｝または｛ｘ^ｏｕｔ（ｎ）｝を相応にシフトさせた後に実行される。

If there is a constant lead or lag of only a few points between the CPR reference signals {x (n)} and {y (n)}, the shift cross-correlation with N _lag / 2 or N _lag points in both segments is specific The maximum value is caused by the lead or lag. The above subtraction is performed after the estimated artifact signal {a ₁ (n)} or {x ^out (n)} is shifted accordingly.

Next, move to a new window display segment {y ₂ (n)} and {x ₂ (n)} by rapidly increasing the N _spike point between the N _spike and the (N _spike + W _L ) samples. The previous step is then repeated using the new window parameters.

  In various embodiments, different weighting methods may be utilized in estimating the second, third, and subsequent non-overlapping segments of data. These weighting methods include equal weighting and center segment weighting.

In the equal weighting method, the overlapping segment between two adjacent windows will be calculated twice, so even weighting is given to the calculation of two adjacent windows if the two windows only overlap. Will be. Similar is the case for three adjacent calculations, and certain sub-segments are common to them. For example to set the _{N lag} / N _jump of 16 points, if having a _{W L} (ie) 128 points, the particular segment can be present in the eight neighboring windows.

First segment of N _spikes or N _lag points:

Ｎ_急上昇またはＮ_ｌａｇポイントの第２セグメント：

Second segment of N _spikes or N _lag points:

Ｎ_急上昇またはＮ_ｌａｇポイントの第３セグメント：

Third segment of N _spikes or N _lag points:

第８セグメントからは、８つの重複窓全てが計算に利用できる。従って、比Ｗ_Ｌ：Ｎ_ｌａｇ＝８：１なら、Ｎ_ｌａｇポイントの各々の重複セグメントは、８回計算される。特定の実施形態において、これらの計算に影響を及ぼすコンボリューションのために、精度に負の影響を及ぼす終端効果の可能性が存在する。従って、実施形態は、窓、Ｗ_Ｌでの結果から終端セグメントを除去する。その場合、重複セグメントは、終端に位置する。

From the 8th segment, all 8 overlapping windows are available for calculation. Thus, if the ratio W _L : N _lag = 8: 1, each overlapping segment of N _lag points is calculated 8 times. In certain embodiments, because of the convolution that affects these calculations, there is a possibility of termination effects that negatively affect accuracy. Thus, embodiments remove the end segment window, the results for W _L. In that case, the overlapping segment is located at the end.

Central heavy weighting in the calculation of segments common to adjacent windows is utilized as the weighting method in various embodiments of the piecewise stitching adaptation algorithm. Central heavy weighting method removes a set of windows, the uncertainty provided by the terminal in the convolution of W _L by weighting the center segments. For example, if only six windows are utilized and a particular segment is in the middle, center weighting is to apply weights to the four central windows, avoiding the windows on the two ends. The center weighting algorithm is represented as follows:

中央重付けの結果として、種々の実施形態は、コンボリューションにおいてより安定性を示す。

As a result of center weighting, the various embodiments are more stable in convolution.

  With reference to FIG. 4, a system for implementing a piecewise stitching adaptation algorithm using a convolution algorithm according to one embodiment is presented. The first select sample window is selected from the primary and secondary observation signals 400. Next, a primary signal segment and a secondary signal segment are generated (step S402). Next, autocorrelation and cross-correlation are calculated from the primary signal segment and the secondary signal segment (step S404). Next, an impulse response is measured from the autocorrelation and the cross correlation (step S406). Next, the estimated artifact existing in the primary signal is estimated from the impulse response (step S408). The estimated artifact is then shortened to match the lengths of the selected primary signal sequence and secondary signal sequence (step S410). Next, the DC response of the estimated artifact is integrated using convolution (step S412). The estimated clean primary signal is generated by removing estimated artifacts from the selected primary signal 414. At this point the cycle may be repeated for the next selected primary and secondary signal sample windows. In this way, embodiments may filter the entire primary signal with high accuracy by utilizing a piecewise stitching adaptation algorithm and a convolution algorithm.

  Referring now to FIG. 5, an embodiment of a piecewise stitching adaptation algorithm that utilizes overlapping regression segments is shown. This embodiment is similar to a window-based overlapping convolution output segment, except that the output segment is calculated from a simple regression relationship between {x (n)} and {y (n)}. . In this case, it is assumed that the artifact and the reference signal have a correlation one by one, even if the relationship is obscured by other components. That is, the relationship is assumed to be piecewise linear.

In this embodiment, α and β are estimated every W _L sample window (˜2 seconds), where the window is the overlap of a given W _L / 2 sample. Nonstationarity is explained to some extent by this overlap. In each window, α ^ and β ^ are calculated by equation (30). Hereinafter, “α ^” indicates a symbol “^” on α.

a ₁ ′ (n) = α ^ x (n) + β ^ (31)
.

Clean signal estimation = y (n) −a ₁ ′ (n) (32)
In this case, since the linear regression equation fits between {y (n)} and {x (n)}, the artifact component is estimated, and the estimated artifact component is obtained from the observed ECG signal {y (n)}. Removed.

  The estimation of α ^ and β ^ in equation (30) can help to reconstruct the artifact components in the observed corrupted ECG signal {y (n)}. In addition, additional windowing and overlapping methods can be used to simulate overlapping convolution methods and can eliminate non-stationarity problems. By using even smaller segments, non-stationary relationships can be captured.

  Referring to FIG. 6, an embodiment of a piecewise stitching adaptation algorithm that utilizes linear regression segments is shown. First, a selected sample window is selected from primary and secondary observation signals (step S600). Next, a primary signal number sequence and a secondary signal number sequence are generated (step S602). Next, linear regression is measured between the primary signal and the secondary signal (step S604). Next, an estimated artifact is generated based on linear regression (step S606). Next, the estimated artifact is shortened to match the lengths of the selected primary signal sequence and secondary signal sequence (step S608). The estimated clean primary signal is generated by removing estimated artifacts from the selected primary signal (step S610). At this point, the cycle may be repeated for the next selected primary and secondary signal sample windows. In this way, various embodiments may filter the entire primary signal with high accuracy by utilizing a piecewise stitching adaptation algorithm and a linear regression algorithm.

  As shown in FIG. 7, the time-varying adaptive windowing method is also enabled by implementing piecewise regression of the piecewise stitching adaptation algorithm method. In this setting, the overlap segment size, the window start point at {x (n)} and {y (n)}, can all vary with the size of the window. This method provides segmental correlation, where segments are biased because they are not the same and the number of calculations per point is different. However, since an acceptable artifact waveform cannot be constructed, it can be subtracted from {y (n)} to derive an artifact removal ECG signal.

  In various embodiments, all signal samples in both the sensed ECG signal and the CPR reference signal initiate calculation and parameter storage operations. For example, the calculation of the average value, the calculation of the autocorrelation sequence, the calculation of the cross-correlation sequence, and the operations of deconvolution and convolution occur continuously. However, in certain embodiments, such as additional calculations such as estimation, the artifact segment is only performed with a specific index. In this way, the piecewise stitching adaptation algorithm can reduce the amount of computation required in signal filtering and analysis.

  With reference to FIGS. 8A-8D, various embodiments of systems utilizing a piecewise stitching adaptation algorithm are shown. The embodiment shown in FIG. 8A inputs an ECG signal 800 and a CPR reference signal 802 into a piecewise stitching adaptation algorithm 804. The piece stitching adaptation algorithm 804 will then remove CPR artifacts from the ECG signal 800 and provide the estimated actual ECG signal 806 for further use.

Referring to FIG. 8B, an embodiment of a system that utilizes a piecewise stitching adaptation algorithm in the absence of a reference signal is shown. In many cases, the low frequency domain ECG signal 800 recorded during CPR is dominated by CPR artifacts. Thus, in the absence of a reference signal, filtering of the ECG signal 800 using the low pass filter 808 enables the features of the reference signal. Although not as accurate as the exact CPR related signal, the output signal 810 processed by this low pass filter 808 can be used as {x (n)} in the piecewise stitching adaptation algorithm 804. In various embodiments, the processing delay is minimized or linearized by using a finite impulse response (FIR) filter. Lead or lag calculations can be applied subsequently using correlation analysis. In various embodiments, the low pass filter 808 filters a select signal range of, for example, (0-6) Hz.

Referring to FIG. 8C, various embodiments employ a piecewise stitching adaptation algorithm 804 to clean CPR artifacts from some input signal artifacts from some input signals by using a CPR reference signal 802. Can be used. Input signals may include an ECG signal 800, a pO ₂ signal 812, an atrial blood pressure (ABP) signal 814, a central venous pressure (CVP) signal 816, and a heart sound signal 818. In this way, the piecewise stitching adaptation algorithm 804 can simultaneously filter CPR artifacts from multiple signals and provide uncorrupted signals for further analysis.

  Referring to FIG. 8D, an embodiment is shown that utilizes a piecewise stitching adaptation algorithm for multiple signals having associated components in the absence of the CPR reference signal 802. If the CPR reference signal 802 is not available, two or more signals corrupted by motion artifacts as a result of CPR are fed into a piecewise stitching adaptation algorithm 804. For example, ECG signal 800 and hemodynamic signal 820 may be input to segment stitching adaptation algorithm 804. In these embodiments, the piecewise stitching adaptation algorithm 804 can correlate the signals and estimate the CPR artifact component. Next, the segmented stitching adaptation algorithm 804 will remove the CPR artifact component from the original source signal and further output a clean input signal 822 along with a signal indicating an estimate of the CPR artifact 824.

  As shown in FIG. 9, in various embodiments, the segmented stitching adaptation algorithm may also be utilized in an AED system to enable a first responder command based on the state of the sense signal. For example, the segmented stitching adaptation algorithm 900 can be utilized in conjunction with the rhythm analysis algorithm 902 to determine between the pulseless rhythm 904, the shock rhythm 906, and the non-shock rhythm 908.

  In various embodiments, a pulseless rhythm 904 that requires continuous CPR can be determined in an AED that utilizes a piecewise stitching adaptation algorithm 900. In these situations, the piecewise stitching adaptation algorithm 900 continuously monitors the ECG signal 910 and the CPR reference signal 912 and sends a clean ECG output signal 914 to the rhythm analysis algorithm 902 that is also implemented in the AED. The rhythm analysis algorithm 902 may then determine that continuous CPR is essential for pulseless electrical activity or cardiac arrest and may cause the AED to deliver a continuous CPR command 916 to the first responder.

Stopping CPR and applying shocks in the presence of a shocking rhythm that requires an existing CPR is done in near real time in an AED that uses a piecewise stitching adaptation algorithm 900. In these circumstances, the piecewise stitching adaptation algorithm 900 continuously monitors the ECG signal 910 and the CPR reference signal 912 and sends a clean ECG output signal 914 to the rhythm analysis algorithm 902. The rhythm analysis algorithm 902 can then determine that an electric shock is needed and the protocol will change the CPR.
The AED may be instructed to deliver a stop command and a shock command 918 to the patient. In this way, the segment stitching adaptation algorithm 900 may interrupt CPR to provide therapy according to the protocol. Various embodiments allow a control signal to be directed to a mechanical compression device that delivers automatic CPR compression to automatically synchronize the cycle of electric shock and compression. If a non-shock rhythm is present, the AED may deliver a CPR continuation command and other commands 920.

  In certain embodiments, the AED utilizes a piecewise stitching adaptation algorithm 900 to determine which signal processing is required by the piecewise stitching adaptation algorithm 900. In these embodiments, the piecewise stitching adaptation algorithm 900 outputs a sensed ECG signal in situations where the estimated artifact signal 916 exhibits a low signal amplitude or where the artifact has only a minor effect on the ECG signal 910. . Thus, in these situations, the sensed ECG signal 910 can be utilized by ignoring the piecewise stitching adaptation algorithm 900 signal processing and can be sent directly to the rhythm analysis algorithm 902, further reducing latency and processing power requirements. .

  By implementing a piecewise stitching adaptation algorithm and utilizing CPR reference symbols, various embodiments can remove noise from the sense signal if the signal-to-noise ratio exhibits very large variations. For example, a piecewise stitching adaptation algorithm can efficiently depict artifacts even for severe artifact conditions with a signal-to-noise ratio of 1-20, and in the case of cardiac arrest this ratio is approximately 1 : Can be much larger than 1000, so is limited primarily by the inherent noise characteristics of the ECG channel.

  The sensing signal as shown in FIG. 10A and the CPR reference signal as shown in FIG. 10B can be analyzed by the piecewise stitching adaptation algorithm in various embodiments, and CPR artifacts can be separated, thereby resulting in FIG. A recovered ECG signal as shown in FIG. As described above, in various other embodiments, the piecewise stitching adaptation algorithm may perform the same analysis on different various primary and secondary signals for the same effect. Thus, for example, various signals that interfere with the load ECG tester and the EEG tester can be eliminated by utilizing a piecewise stitching adaptation algorithm.

  Referring to FIG. 11, a block diagram of an AED 1000 that implements a piecewise stitching adaptation algorithm according to one embodiment of the present invention is shown. A digital microprocessor assisted control system 1002 is used to control the overall operation of the AED 1000. The electric control system 1000 further includes an impedance measurement circuit that inspects the interconnection and operability of the first electrode 1004 and the second electrode 1006. The control system 1002 has a processor 1008 interfaced to a program memory 1010, a data memory 1012, an event memory 1014, and a real time clock 1016. The operation program executed by the processor 1008 is stored in the program memory 1010. The electrical output is provided by the battery 1018 and connected to the generator circuit 1020.

The power generation circuit 1020 is also connected to an output adjustment device 1022, a lid switch 1024, a supervisor timer 1026, a real time clock 1016, and a processor 1008. Data communication port 1028 is coupled to processor 1008 for data transfer. In certain embodiments, data transfer may be performed by utilizing a serial port, USB port, firewire, eg, 802.11x or 3G, radio such as radio. Rescue switch 1030, maintenance indicator 1032, diagnostic display panel 1034, audio circuit 1036, and alarm device 1038 are similarly connected to processor 1008. The audio circuit 1036 is connected to the speaker 1040. In various embodiments, the rescue light switch 1042 and the visual display 1044 may be used by the processor 100 to provide additional driving information.
8 is connected.

  In certain embodiments, the AED may include a processor 1008 and a piecewise stitching adaptation algorithm co-processor 1046. The piece stitching adaptation algorithm coprocessor 1046 may be a piecewise stitching adaptation algorithm implemented in hardware and may be operatively connected to the processor via a high speed data bus. In various embodiments, the processor 1018 and the piece stitching adaptation algorithm coprocessor 1046 are on the same silicon and may be implemented in a multi-core processor. The processor 1008 and the piecewise stitching adaptation algorithm coprocessor can also be implemented as part of a multiprocessor or as a network processor arrangement. In these embodiments, the processor 1018 offloads a portion of the piecewise stitching adaptation algorithm calculation to the piecewise stitching adaptation algorithm coprocessor, thus optimizing the processing of the detection signals from the first electrode 1004 and the second electrode 1006. Turn into. In other embodiments, the processor 1008 is optimized with special instructions or optimizations to perform piecewise stitching adaptation algorithm calculations. Accordingly, the processor 1010 may perform the piece stitching adaptation algorithm calculation in fewer clock cycles while using fewer hardware resources. In other embodiments, the logic and algorithms of the control system 1002 may be implemented with ASIC type hardware, FPGA type combinations, or any other logic.

  High voltage generation circuit 1048 is also connected to and controlled by processor 1008. The high voltage generation circuit 1048 may include a semiconductor switch (not shown) and a plurality of capacitors (not shown). In various embodiments, connectors 1050, 1052 connect high voltage generation circuit 1048 to first electrode 1004 and second electrode 1006.

  Impedance measurement circuit 1054 is connected to both connector 1050 and real time clock 1016. The impedance measurement circuit 1054 is interfaced to a real time clock through an A / D (analog / digital) converter 1056. Another impedance measurement circuit 1058 may be connected to the connector 1050 and the real time clock 1016 and may be interfaced to the processor 1008 through the A / D converter 1056. CPR device 1060 may be connected to processor 1008 and real time clock 1016 through connector 1052 and A / D converter 1056. The CPR device 1060 can be a chest compression detection device or a manual, automatic or semi-automatic mechanical chest compression device.

  Similarly, it is to be understood that the exemplary embodiment or exemplary embodiments are only examples, and are not intended to limit the scope, applicability, or configuration of the invention in any way. Rather, the foregoing detailed description will provide those skilled in the art with disclosure that will help them to implement the exemplary embodiment or embodiments. It will be appreciated that various changes may be made in the function and arrangement of elements without departing from the scope of the invention as set forth in the appended claims and their legal equivalents. .

  The above embodiments are intended to be illustrative rather than limiting. Additional embodiments are within the scope of the claims. Furthermore, while features of the invention have been described with reference to specific embodiments, those skilled in the art will recognize the form and details without departing from the spirit and scope of the invention as defined by the claims. Will admit that it can be changed.

Those skilled in the art will appreciate that the present invention may have fewer features than are exemplified in any individual embodiment described above. The embodiments described herein are not exhaustive of the ways in which the various features of the present invention can be combined. Thus, the embodiments are not mutually exclusive feature combinations, but rather the invention may have different individual feature combinations selected from different individual embodiments, as will be understood by those skilled in the art. .

  Any incorporation by reference of documents above is limited such that no subject matter is contrary to the express disclosure herein. Any incorporation by reference of documents above is further limited such that claims contained in the document are not incorporated by reference herein. Any incorporation by reference of documents above is further limited so that any definition provided in the document is not incorporated by reference herein unless expressly included herein.

  In order to be able to interpret the claims of the present invention, the special terms “means for” or “steps for” are subject to the provisions of section 212, paragraph 6 of the 35th Code of the United States, unless they are listed in the claims. It is intended to clarify what should not be done.

Claims (15)

A filter device for filtering artifacts from an ECG signal in real time, the filter device comprising:
Means for detecting an ECG signal representative of a physical impulse of cardiac tissue;
Means for detecting an artifact signal representing a physiological function;
Means for removing the artifact signal from the ECG signal by using a piecewise stitching adaptation algorithm ;
The segment stitching adaptation algorithm is:
Means for selecting a plurality of signal sample windows from the ECG signal and the artifact signal;
Means for generating a primary ECG signal segment from the ECG signal based on the plurality of selected signal sample windows and generating a primary artifact signal segment from the artifact signal;
Means for measuring a relationship between the primary ECG signal segment and the primary artifact segment;
Means for estimating an ECG signal artifact in the primary ECG signal segment based on the measured relationship;
Means for removing the estimated ECG signal artifact from the primary ECG signal segment;
It comprises, filter device. A filter device having a processor configured to filter CPR compression artifacts from an ECG signal in real time, the filter device comprising:
An ECG sensor for detecting an ECG signal representative of a physical impulse of cardiac tissue;
An artifact sensor for detecting an artifact signal representing a physiological function;
The ECG sensor and the artifact programmed to calculate an ECG signal artifact caused by the artifact signal and to remove the ECG signal artifact from the ECG signal using a piecewise stitching adaptation algorithm A segmented stitching adaptation algorithm processor coupled to the sensor ;
The piece stitching adaptation algorithm processor is configured to execute instructions stored in a memory operatively coupled to the piece stitching adaptation algorithm processor, the instructions comprising:
Selecting a plurality of signal sample windows from the ECG signal and the artifact signal;
Generating a primary ECG signal segment from the ECG signal and generating a primary artifact signal segment from the artifact signal;
Measuring a relationship between the primary ECG signal segment and the primary artifact signal segment;
Estimating a signal artifact in the primary signal based on the measured relationship;
Removing the estimated signal artifact from the primary ECG signal segment;
The having, filter apparatus. A method for operating a filter device as a machine-implemented process for filtering signal artifacts from ECG signals in real time, comprising:
The filter device includes:
An ECG sensor for sensing a ECG signal representing a physical impulse of cardiac tissue;
And artifact sensor for sensing an artifact signal representing a physiological function;
A segmented stitching adaptation algorithm processor coupled to the ECG sensor and the artifact sensor;
Have
The method of operation, in order to automatically remove the signal artifact from the ECG signal have use the PSAA to generate a clean ECG signal, the PSAA processor performs the steps below Prepared
Selecting a first signal sample window from the ECG signal and selecting a second signal sample window from the artifact signal;
Generating a primary ECG signal segment from the first signal sample window and generating a primary artifact signal segment from the second signal sample window;
Measuring a relationship between the primary ECG signal segment and the primary artifact signal segment;
Estimating signal artifacts in the primary ECG signal segment based on the relationship;
A method of operating a filter device, wherein the estimated signal artifact is removed from the primary ECG signal segment . The method of operation further comprises using a rhythm analysis algorithm processor to identify a shocking ECG rhythm.
The operating method according to claim 3 . When the artifact sensor detects an artifact representing a physical impulse,
In order to automatically remove the signal artifact PSAA from the ECG signal, it activates the PSAA processor,
The operating method according to claim 3 . Selecting the first signal sample window from the ECG signal and selecting the second signal sample window from the artifact signal;
Further comprising selecting a signal sample window selected from the group consisting of signal sample windows of uniform size and non-uniform size depending on a time delay between the ECG signal and the artifact signal.
The operating method according to claim 3 . Selecting the first signal sample window from the ECG signal and selecting the second signal sample window from the artifact signal;
Furthermore with reference to the end time and integrity of the start time, comprises selecting a said first signal sample window and the second signal sample window,
The operating method according to claim 3 . Selecting the first signal sample window from the ECG signal and selecting the second signal sample window from the artifact signal;
Using signal sample window start time of the non-integrity and signal sample window end time comprises selecting a said first signal sample window and the second signal sample window,
The operating method according to claim 3 . The operating method further comprises:
Estimating a phase lead or phase lag between the ECG signal and the artifact signal using a shift autocorrelation calculation;
The calculation of the phase lead or phase lag comprises being stored in memory to select a further signal sample window;
The operating method according to claim 3 . The operating method further comprises:
Weighting the primary and secondary signal segments with a weighting method selected from the group consisting of equal weighting and center segment weighting;
The operating method according to claim 3 . The artifact signal is selected from the group consisting of a CPR compression signal and a hemodynamic signal;
The operating method according to claim 3 . The artifact signal is generated by applying a bandpass filter to the ECG signal;
The operating method according to claim 3 . The artifact signal is graded by using time domain estimation to generate a grade,
The time domain estimate is selected from the group consisting of zero crossings and peak-peak oscillations;
The operating method according to claim 3 . The grade indicates the quality of the signal-to-noise ratio and provides a further rhythm identification confidence criterion,
The operating method according to claim 13, wherein. The method of operation further comprises denoising the ECG signal and the artifact signal.
The operating method according to claim 3 .

Images